基于深度学习CenterPoint的3D目标检测部署实战

1.所用硬件

硬件：英伟达 Jetson AGX

环境：ubuntu 20.04

2.安装依赖

yaml 复制代码

CUDA Toolkit: 11.4
python: 3.8
numpy: 1.23.1
pytorch: 2.0.0
ros: notic
rospkg
ros_numpy:  (sudo apt-get install ros-$ros_release-ros-numpy)
pyyaml
argparse

2.拉取代码

bash 复制代码

git clone https://gitee.com/uuuu_qqq/livox_detection.git

cd livox_detection

python3 setup.py develop

3.解决部署过程遇到的bug

3.1 EasyInstallDeprecationWarning: easy_install command is deprecated.

解决EasyInstallDeprecationWarning: easy_install command is deprecated.

3.2 torch cuda版本安装

NVIDIA 优化框架：Jetson 平台 PyTorch 安装指南

3. 3 ModuleNotFoundError: No module named 'ros_numpy'

sudo apt-get install ros-noetic-ros-numpy

遇到下面问题：

bash 复制代码

http://mirrors.tuna.tsinghua.edu.cn/ros/ubuntu/pool/main/r/ros-noetic-ros-numpy/ros-noetic-ros-numpy_0.0.5-2focal.20250510.062009_arm64.deb

E: Unable to fetch some archives, maybe run apt-get update or try with --fix-missing?

于是直接复制下载链接，到官网下载，然后本地安装：

sudo dpkg -i ros-noetic-ros-numpy_0.0.5-xxxxx_arm64.deb

3. 4 AttributeError: module 'numpy' has no attribute 'float'.

bash 复制代码

raise AttributeError(__former_attrs__[attr])
AttributeError: module 'numpy' has no attribute 'float'.
`np.float` was a deprecated alias for the builtin `float`. To avoid this error in existing code, use `float` by itself. Doing this will not modify any behavior and is safe. If you specifically wanted the numpy scalar type, use `np.float64` here.

只需要找到对应代码，把np.float改为float或者np.int改为int;

4.开始执行

运行 ROS.
bash 复制代码
```
roscore
```
进入 'tools' 目录并运行 test_ros.py（预训练模型：.../pt/livox_model_1.pt 或 .../pt/livox_model_2.pt）
bash 复制代码
```
cd tools
python3 test_ros.py --pt ../pt/livox_model_1.pt
```
播放rosbag。 (pointcloud 2的主题应该是/livox/lidar，或者自己在代码里改一下，直接接雷达数据也可以，话题类型要一致)
bash 复制代码
```
rosbag play [bag path]
```
rviz可视化.
bash 复制代码
```
rviz -d rviz.rviz
```